Towards Initialization-Agnostic Clustering with Iterative Adaptive Resonance Theory

TL;DR

IR-ART addresses the sensitivity of Fuzzy ART to the vigilance parameter $\rho$ by embedding three adaptive phases—Cluster Stability Detection, Unstable Clusters Deletion, and Vigilance Region Expansion—into an iterative framework that preserves the original simplicity of Fuzzy ART. The method analyzes iterative changes in cluster structure to identify unstable clusters, removes them, and softly expands the vigilance region for stable clusters, enabling robust clustering across a wide range of $\rho$ without extra hyperparameters. Empirical results across 15 datasets show IR-ART often achieves higher peak and mean clustering performance with lower sensitivity to $\rho$ than competing ART-based methods, though some complex cluster shapes can still pose challenges. The approach offers practical benefits for non-expert users, especially in resource-constrained settings, and suggests potential for integration with other ART-based or federated learning frameworks.

Abstract

The clustering performance of Fuzzy Adaptive Resonance Theory (Fuzzy ART) is highly dependent on the preset vigilance parameter, where deviations in its value can lead to significant fluctuations in clustering results, severely limiting its practicality for non-expert users. Existing approaches generally enhance vigilance parameter robustness through adaptive mechanisms such as particle swarm optimization and fuzzy logic rules. However, they often introduce additional hyperparameters or complex frameworks that contradict the original simplicity of the algorithm. To address this, we propose Iterative Refinement Adaptive Resonance Theory (IR-ART), which integrates three key phases into a unified iterative framework: (1) Cluster Stability Detection: A dynamic stability detection module that identifies unstable clusters by analyzing the change of sample size (number of samples in the cluster) in iteration. (2) Unstable Cluster Deletion: An evolutionary pruning module that eliminates low-quality clusters. (3) Vigilance Region Expansion: A vigilance region expansion mechanism that adaptively adjusts similarity thresholds. Independent of the specific execution of clustering, these three phases sequentially focus on analyzing the implicit knowledge within the iterative process, adjusting weights and vigilance parameters, thereby laying a foundation for the next iteration. Experimental evaluation on 15 datasets demonstrates that IR-ART improves tolerance to suboptimal vigilance parameter values while preserving the parameter simplicity of Fuzzy ART. Case studies visually confirm the algorithm's self-optimization capability through iterative refinement, making it particularly suitable for non-expert users in resource-constrained scenarios.

Figures (4)

• Figure 1: The upper part of this figure briefly presents the overall framework of IR-ART and the Iterations module is detailed in the lower part of this figure. Phases (A) and (B) in the Iterations are similar to those in traditional Fuzzy ART iterations, whereas the core phases of IR-ART are (C), (D), and (E).
• Figure 2: Average NMI during $\rho$ scan on eight datasets for Fuzzy ART, AM-ART, SA-ART, and IR-ART.
• Figure 3: Average number of clusters during $\rho$ scan.
• Figure 4: Clustering results on Flag dataset during the IR-ART's iteration process.